Abstract
This paper provides a practical procedure for designing efficient taxi pick-up operations at airports.
How to do this effectively is an open question. Solutions are not available. Practices differ widely,
reflecting different approaches to, and lack of research on the subject. The solutions are often
unsatisfactory. At many airports, passengers routinely suffer long waits outside, exposed to the
elements, after a tiring journey. Such disagreeable experiences are avoidable.
Designing efficient taxi pick-up operations at airports is problematic. The peculiarities of this
process preclude easy solutions. First, it involves queuing, so system performance is a highly
non-linear function of the loads. Second, it features unstable transient situations as travelers
typically arrive in bulk over short periods. Third, traffic is significantly differentiated, consisting of
a wide variety of groups implying different service characteristics. Standard results from queuing
theory thus do not apply usefully to this problem.
The design process therefore uses simulation based on detailed observation of the local practices.
It involves 4 steps: (1) detailed local measurements of both the arrival of travelers, and the
service rates provided by taxis in different queuing positions; (2) creation and validation of a
simulation model sufficiently detailed to account for these realities; (3) exploration of design
alternatives to estimate the characteristics of the service that they would provide; and (4)
selection of a preferred design that properly balances efforts to minimize average and extreme
wait times. The paper demonstrates this procedure through application to the Lisbon Airport.
Key Words: Airports, Taxis, Terminal Design, Simulation, Queuing Theory

The Problem

Taxi pick-up operations at airports are sources of frequent complaints. Passengers seeking to
get taxis to complete their journeys routinely have long waits outside, exposed to the elements,
after a tiring journey. For example, international passengers arriving in Boston wait outside, in
eventual rain, wind and snow, for taxis to arrive sporadically from a holding lot about 2 kilometers
away. Even when taxis are available, travelers cannot board them rapidly due to the design of
the boarding area. Similarly, passengers arriving in Lisbon can easily queue up for 20 minutes
even though the taxi pool is full – and only 100 meters away.

The problem of designing efficient taxi pick-up operations is important at the many airports where
taxis service a significant fraction of the arriving passengers. Taxi operations are most important
for airports close to their cities, particularly where public transit is either minimally available or
provides poor service to travelers encumbered with bags. La Croix (1991) and Cardon (2007)
have documented this logical conclusion. Boston and Lisbon are good examples of this situation,
along with Bangkok, Chicago, Delhi, Frankfurt, Los Angeles, New York, Paris, San Francisco,
Singapore, Sydney, Toronto and many others. Table 1 illustrates the importance of taxi
operations at selected airports.

No coherent guidance is available to airport operators for how to design efficient taxi pick-up
operations, despite the importance and prevalence of the issue. Essentially no information is
available on the topic. As a practical matter, both the international standard design manual (IATA,
2004) and the major current textbooks (Ashford et al, 2011; Ashford and Moore, 1996; de
Neufville and Odoni, 2003; Horonjett, 2010; Kazda and Caves, 2007) are silent on the issue. The
references to taxi services are limited to a general view of curbside access and taxi availability,
from nearby segregated “pools”. These sources neither address the efficient design of taxi
service, of passenger waiting areas and their interface, nor provide guidance on how to handle
the mismatch between their distinct operational behaviors.

Table 1 – Taxis serve a significant portion of passengers at many airports

In this context, this paper proposes a comprehensive, coherent, and practical approach for
designing efficient taxi operations. It comprehensively deals with the entire problem, from
collecting data, through the analysis, to presenting results meaningfully for airport decisionmakers. It is coherent in that it consistently deals with the reality of transient queuing processes.

As demonstrated by the application to a major airport, the method is practical. Specifically, the
method provides a process for:
• Gathering the kind of data needed the characterize taxi pick-up operations;
• Calibrating the data to peculiarities of local situation;
• Extrapolating this reality to alternative designs;
• Presenting the results showing the impact of alternative designs according to the range of
significant metrics (such as average and maximal wait times, and the reliability of the
process).

The proposed method thus provides airport managers with the kind of information they need to
make informed decisions about how to best suit their taxi operations to local conditions.

Current Practice
Given the lack of detailed guidance, it is not surprising that airports have adopted a hodge-podge
of designs and practices for their taxi operations. As regards the taxis, airports variously line up
them in single, dual or even triple files for taxi queues in front of passengers. Single file
arrangements are perhaps most common. Dual files, as practiced at Washington/Reagan and Lisbon for example, have the advantage of offering the waiting passengers more taxis close to
the head of their queue. Some airports, such as Singapore, use parallel loading, which permits
taxis to exit their queue as soon as they are ready, without waiting for other taxis to complete their
loading process. Figure 1 illustrates these alternatives.
Airport practices also differ in how they treat the passenger side of the taxi pick-up process.
Some allow passengers to board from multiple points; others force them to pass through a single
point of access. Some airports let passengers go directly to available taxis as they arrive, others –
– such as Boston, constrain passengers to wait behind a barrier until a new batch of taxis is in
place.

In short, there is no agreement on how to design and operate taxi pick-up operations. While we
should expect that different solutions are preferable for different situations, it is safe to say that
prevailing practices typically are historical legacies rather than the result of conscious analysis
and consideration.

Nature of the Problem
The problem of designing efficient taxi pick-up operations at airports is technically challenging.
The analysis is conceptually difficult and the structure of the problem does not allow for closed
form solutions. These facts go a long way to explaining why general solutions are not available.
Understanding the nature of these difficulties points the way to developing an effective procedure
for defining efficient designs.

The analysis of taxi pick-up operations is conceptually difficult because it involves queuing. That
is, people arrive looking for service [to board a taxi], they queue up behind the position(s) at
which service is offered, and then exit the process [when they leave in a taxi]. The essential
difficulty with a queuing problem is that its performance depends significantly and non-linearly on
the state of the system.

In the first instance, the level of delays and the length of the queues depend greatly on the
relative load on the system. Both the delays and the unreliability of the system in steady state
increase rapidly as the rate of arrivals nears its nominal capacity (Figure 2a). Further, this nonlinearity increases as the rate of arrivals at the point of service becomes more irregular (Figure
2b) and as defined by Equation 1 and 2 for a Poisson arrivals process.

This phenomenon leads counter-intuitively to the fact that the “practical capacity” of the system –
that we can think of as the quantity of arrivals that could be served within acceptable delays and
levels of reliability – is substantially less than the nominal capacity. Specialists understand this
phenomenon, which provided the basis for the old practice of defining the capacity of runways in
terms of their ability to handle arrivals with less than a stated number of minutes of delay
(Horonjeff, 1962). Yet managers often miss this reality as they wrongly assume that a service
system will operate satisfactorily up to its nominal capacity. Thus the designers for the
automated baggage handling system for Denver International Airport failed to account for this and
were dismayed to find that their system became too unreliable – and thus effectively at “practical
capacity” — whenever loads exceeded around 40% of its rated capacity.(de Neufville, 1994).

The analysis of taxi pick-up operations is further complicated because it involves transient loads,
that is, a fluctuating rate of arrivals for service. The rate at which travelers arrive at taxi stands
typically varies significantly over the course of any hour and any day, as various flights arrive.
Periods of intense demand for taxis follow long slack periods. The resulting queuing process is
thus rarely in steady state. There is then no closed-form description of the queuing process of
the kind provided in Equation 1. Theoretical queuing analysis thus does not apply in useful detail.
Finally, the analysis is difficult because the stream of passengers arriving for taxi service is highly
differentiated. Customers arrive in groups of various sizes and capabilities to deal with the
process of boarding taxis. For example, a single business traveler coming to town for a day may
hop into a taxi and move on quickly. In contrast, the family of four foreigners arriving for a
vacation will require much more time as the load their many bags and try to explain their
destination to the driver. Thus, the rate at which the taxi pick-up system can provide service is
also a mixed bag of processing times. In this respect queues for taxi operations are quite distinct
from many other queuing operations in airports, in which travelers come in groups but are
processed as individuals, for example at security check points.

Additionally, the rate at which individual taxis can provide service depends on their position in the
queue. Specifically, the rate at which the second or third taxi in a line can complete service [as
by leaving the queue and leaving space for a replacement] usually depends on the speed at
which the taxi(s) ahead of it provide service. This last point is particularly important for the design
of taxi pick-up operations. Alternative designs for these processes position taxis differently, as
Figure 1 shows. In order to evaluate the performance of design alternatives, we therefore need
to have a method of analysis that is sensitive to the positions of the taxis in the alternative
designs for the operations.

In summary, the nature of taxi pick-up operations at airports precludes theoretic, analytic
solutions. Furthermore, as the processing rate of the pick-up operation depends greatly both on
the kind of passenger group being served and the position of a taxi in the queue, we need a
method of analysis that is sensitive to such details. A general, all-purpose model will not do.
Analysts therefore depend on some form of detailed simulation carefully calibrated to the realities
of a specific situation. This is the approach presented in this paper.

Review of Literature
This section reviews the elements relevant to the design of taxi pick-up operations at airports:
1. analyses of taxi pick-up operations;
2. practical guidance about managing queues;
3. approaches to dealing with transient queues; and
4. pertinent topics in simulation.
Analyses of Taxi Pick-up Operations:
To our knowledge, there are only four reports on the subject within the open domain. Notably, as
indicated in the introduction, none of the standard textbooks or manuals on airport design and
operations analyze taxi operations.

The first is that of Lee (1966) who presented a detailed analysis of the West London Air Terminal.
His work stresses the necessity to examine carefully the behavior of arrivals and service time, in
particular the distribution of groups of passengers that translate into single taxi hires and trips.
Curry and De Vany (1977) modeled the transient buildup and dissipation of taxi and bus queues
at Dallas/Fort Worth Airport. They approximated the transient levels of arrivals as a series of
steady-state cases, which they used to calculate expected wait times and other measures of
performance. They did not suggest a general procedure for analyzing alternative ways to design
the taxi operations.

Costa (2009) analyzed the taxi operations in detail at Lisbon International Airport. This paper
reports on these results, as extended by his subsequent work as an employee of the operator of
Lisbon and other Portuguese airports, ANA, Aeroportos de Portugal, SA.
Finally, based upon Costa’s work, Ferreira, Pacheco and Teodoro (2010) tested a specific
proposed alternative design of the taxi operations at Lisbon International Airport though means of
closed-form formulaic approximations for important system characteristics.
Practical Guidance about Managing Queues

For designing taxi pick-up operations, perhaps the most important guidance from queuing theory
is the idea that operators can improve overall service by providing differentiated service to
customers with distinct service times. This is the principle behind providing express lanes for
customers who can be served quickly at checkout counters for grocery stores and the like. These
fast lanes expedite service and shorten the wait time and queue length for the overall process.
The concept of fast lanes opens up the possibility of better service in situations where operators
could implement the equivalent of fast lanes. These might occur at airports with large traffic at
where it might be sensible to have multiple pick-up points, separating destinations, as between
central business areas and suburban destinations for example. This approach is common
practice for regional bus service in many hub airports, notably in Asia at Tokyo/Narita and
Seoul/Incheon. In these cases, passengers queue up in many different positions along the curb,
each assigned to specific regional destinations.

It is also worth mention, because it analysts often forget, that small seemingly trivial changes in
processing arrangements can have substantial effects on overall performance. Saving 15
seconds per operation might seem insignificant, but when these operations take two to three
minutes to execute – as typical of many airport services and of taxi pick-ups in particular—such
savings can increase total capacity by 10% or more.
Particularly relevant here is the practice of pre-positioning customers from a long queue in front of
the place where they will receive service. At busy airports of entry, the US Immigration services
thus now direct international arrivals, who start out in a single snake queue, into secondary
queues just in front of the kiosk and officer who will process their documents. This practice
minimizes the time between when one customer exits service and the next appears. It contrasts
with the more general situation, as at check-in counters, where customers may not leave the
head of the snake queue until they notice that one of the service channels is open. When the open channel is 10 or more meters away, as can easily happen at busy airports, customers will
take time both to notice that the service is available, and to move to the agent providing the
service. The result is that this agent will have more time between customers, will provide lower
capacity and the overall performance of the entire service process will be less than it could be.
The lesson here is that the design of taxi pick-up operations should pay close attention to details
that can expedite service and improve performance.
Newell’s Approach to Dealing with Transient Queues

Newell (1982) developed and presented a practical way to estimate service quality provided by
service systems under transient loads. Briefly, his approach assumes that as traffic builds up it is
served without delay until arriving traffic equals the prevailing capacity of the service system.
Further traffic would be served at the rate of this capacity, and customers in excess of this
capacity would accumulate in a queue and would be served eventually as the number of
customers arriving for service eventually became less then prevailing service capacity of the
system. The concept leads to a simple graphical analysis. Figure 3, based on Costa’s data from
Lisbon illustrates the approach. As indicated, the horizontal distance between the arrivals and
service defines the queue length, and the vertical distance approximates the wait time.

Newell’s approach is quick and convenient. It has the great advantage of providing a means to
estimate the effect of changing service rates, for example by opening or closing check-in counters or security inspection lanes. Its drawback is that it only works with a specific scenario. It thus
does not provide guidance on distribution of wait times or the unreliability of the system. This
approach thus does not help managers understand the “practical capacity” of a particular design
of a service system, or give information about the frequency of unsatisfactory service that may
define the overall acceptability of a design.
Overall, Newell’s approach is perhaps most useful in providing insight into how managers of
operations can affect average service by scheduling the shifts of personnel to provide service
capacity at the right time and place. As airport operators and managers generally have little
control over the availability of taxi service, Newell’s approach does not seem applicable in that
regard to the design of taxi pick-up operations.
Pertinent References in Simulation
Simulation is now widely invoked as a means to study and thus to improve airport operations of
all kinds, both landside and airside. Analysts commonly use it to study various actual or potential
bottlenecks in airport terminals, both in advance of construction and afterwards, once problems
arise. Common applications include check-in processes and security inspection stations.
Mumayiz and Schonfeld (1999) provide a collection of applications, which include those of
Bender and Chang (1999) and Hathaway (1999) for roadway operations.
Generic computer-based simulations are easy to code. This means that there are many versions
available for potential consumers. Generic simulations are however generally of much use in
analyzing operations. This is because the output of plain vanilla simulations can be difficult to
understand or appreciate. For this reason, providers of simulation services generally spend
considerable time to prepare attractive visual images, typically in some form of video, to illustrate
how their simulation works. These graphical capabilities differentiate the simulations in the
marketplace. They are thus proprietary and unavailable for public discussion. Good graphics
help sell simulations to customers. However, they do not themselves make the simulation
analytically valuable.

The fundamental difficulty in applying simulations to people lies in describing the flows correctly.
It is important to understand how these actually occur. They are typically quite complicated.
People are not like parts in an assembly line that flow in straight paths from entry to exit. People
wander around, between processes. Between check-in and the security inspection, they may buy
a magazine or say good-bye to loved ones, for example. Understanding the complexity of these
flows is a crucial part of creating effective simulations. Simulations prepared by persons who are not deeply familiar with airport operations can thus easily be misleading, because they are naïve
about how flows actually occur.
Complementarily, truly effective simulations depend on having good data about the details of the
arrivals. As Lee (1966) stressed long ago, many processes are strongly affected by how
customers arrive in groups. Are they single, or with family or friends? Are they frequent flyers
that know their way around and can proceed expeditiously, or are they encumbered or
proceeding with some difficulty? These factors can significantly affect the overall process. For
this reason, aggregate data, as about the number of customers per hour or the distribution of
arrival times, are not sufficient to define effective simulations of airport operations. This fact
motivates the proposed process for analyzing taxi pick-up operations.
Proposed process
The proposed process for analyzing taxi pick-up operations consists of four steps:
1. Detailed measurement of arrivals (size and number of groups, interarrival times); and of
service characteristics (the way loading times and exits from queue depend on
configuration of queue and position in it);
2. Creation and calibration of simulation model to the specifics of the operation;
3. Exploration of alternatives, by applying the calibrated model to alternative designs; and
4. Choice of preferred design, which comes down to choosing between alternative
distributions, thus involving multiple criteria – not just average wait time, but considering
possible extremes.

1. Detailed measurement of arrivals and service characteristics
Passengers arrive at the taxi pick-up points singly and in groups. Further, the groups do not
arrive regularly, but spread out over time. An obvious first step is thus to observe the sequence
of persons arriving at the taxi stand. In this regard, note that this distribution of groups is almost
certain to be different from the distribution of groups in other operations, such as security
checkpoints. This is because people self-select for taxi service. Groups for example may
preferentially choose taxis, because the cost per person is low relative to that of individual
travelers, thus making taxi service a more attractive value proposition for groups. It is thus
necessary to measure the arrival times and sizes of groups at the taxi stand. This observation
should be straightforward.

Obtaining good data on the service process is more complicated. This is because each position
in a queue of taxis picking up passengers will typically have its own service characteristics.
These depend both on the distance between its position in the queue and the head of the queue  of passengers, which influences the time it takes to board its passenger(s), and on the its position
in the queue, which determines how long it must wait for the taxis ahead of it to leave and thus
permit it to complete the pick-up service. Furthermore, the service time depends on details of
how the taxis can locate themselves in the queue, which depends on the way the lanes are
marked off, the pattern of police or other enforcement, and so on. In general, the process of
collecting data on the taxi-pick up process requires two phases. The first is a period of
observation to identify the local particularities of the process to know what one should be
measuring. That done, analysts can devise an effective plan for measuring the service process.
2. Creation and calibration of simulation model
Any simulation model is an assembly of basic modules, such as those depicting arrivals of
customers for service into a queue, servicing processes, and departures from service to either
some holding area or another queue. Creation of a suitable simulation model for taxi pick-up
operations thus depends on having a good understanding of the nature of the flows at this
location. This understanding is obtained through step 1. Once analysts have this, the creation of
the simulation model is obvious.
The calibration of the model is tricky. As the measurements on the system include the
determining elements of the queuing process (the arrivals and service rates), a correctly
assembled simulation should reproduce observed results such as the queue lengths. However,
this is often not the case. It may easily happen that the initial measurements in step 1 missed
some factor that caused the distortion. Analysts should determine the cause of such
discrepancies by observation and subsequent adjustment of the simulation until it leads to
reasonably close representations of the observed behavior.
3. Exploration of alternatives
Given a simulation model satisfactorily calibrated to local behaviors of passengers and taxi
operators, the analyst is now in position to explore and compare alternative designs of the taxi
pick-up operations. This evaluation process should both consider the range of possible inputs
and outputs.
It is important to consider the prospective performance of the pick-up system both with different
levels of loads and patterns of demand. This is because queuing systems are non-linear, such
that queue lengths and waiting times are likely to increase much faster than the level of loads
(Figure 2). Furthermore, because taxi pick-up operations are subject to fluctuating loads, they
rarely are in steady state: this means that they are sensitive to the precise pattern of arrivals for
service. In short, the pick-up operations may exhibit widely different performance in terms of queue lengths and wait times for the same average hourly or day level of traffic, depending on
whether arrivals spread smoothly over the period or clump together at particular time.
Complementarily, the analysis should recognize that different designs produce different
distributions of performance – some will be more reliable (have lower variance) than others.
These differences matter. Customers are typically bothered both by delays and by their
unpredictability. Therefore, the analyst should consider not only average performances of
alternative designs, but also the range of performance. Note that the notion of a standard
deviation in this context is not especially useful by itself. This is because performance is
asymmetric: the good performance is limited to zero delays and queues, whereas bad
performance can deviate greatly from the average. It is thus more appropriate to refer to
maximum and minimum performances, especially for decision makers who may not appreciate
the use of technical terms.
4. Choice of preferred design
In the context of uncertain outcomes, such as those resulting from queuing systems, it is
generally best to refer to preferred, rather than best designs (de Neufville and Scholtes, 2011).
This is because the measures of good performance and of cost reflect different, incomparable
units. While we can agree that both delays and service unreliability are undesirable, we are
unlikely to agree on how to compare these units – let alone give them a monetary value. Absent
an agreed upon way to weight the different characteristics of any alternative, it is impossible to
define an unambiguous best design. It is thus better to refer to preferred designs, those whose
combinations of attributes stakeholders prefer.
The analysis must thus also recognize that the designs that the decision-makers prefer may differ
substantially from those favored by the analyst. They may be more sensitive to costs, for
example, or particularly responsive to unreliable service. This fact implies that the analyst should
provide the decision-makers with the data on the range of performance for the several
alternatives. The analyst might best do this by using a table as the following application indicates.

Case Application: Lisbon International Airport
Costa (2009) developed and applied the proposed process at the Lisbon International Airport, in
cooperation with the airport operators, ANA. The case demonstrates the usefulness of the
process and reports on the specific findings.
1. Detailed measurement of arrivals and service characteristics
The main taxi pick-up operation at Lisbon International Airport features two rows of taxis, each
having two taxis providing service at any time. Taxis feed into these rows from a single file
constantly replenished from taxi pool about 100m away. Customers line up in a snake queue
whose head is at the top of the taxi row. Figure 4 shows the details.

Following extensive reconnaissance of the operations, Costa (2009) developed the data
collection plan shown in Figure 5. It involves two observers. One monitored the performance of
the loading process, that is, the service. The other observed the arrivals, noting both their arrival
times and the extent to which the customers were part of a group. The observers took detailed
measurements on different days.

Figures 6 to 8 provide the data on the arrivals. Figure 6 shows that about 2/3 of the customers
arrive reasonably close together, and the rest spread out over time. This is consistent with a
Poisson distribution, which might be expected. Figures 7 and 8 indicate that about 3/4 of the
groups consist either of single customers or a couple. Note that while larger groups are relatively
infrequent in this case, together they account for about 40% of the total customers.

Figures 9 and 10 indicate the service times for the taxi pick-up operations. The four histograms in
Figure 9 give the service times associated with each of the four positions indicated in Figure 4.
Servers 1 and 2 are at the front of the two rows, and servers 1 and 3 are closest to the curb and
thus to the customers waiting in the queue. Figure 10 provides the averages and standard
deviations on these data. As expected, the taxis second in line and further out take longer times
– both because their passenger take longer to get to these taxis, especially if they have to cross a
lane of taxis, and because the taxis cannot complete their service and move out until the taxi in
front of them has also completed its service. Figure 11 illustrates the conflicts between taxis and
pedestrians that lead to these degradations in performance.
Figure 9. Histograms for Service Times
Unit Row!A Row!B
Server Server!1 Server!3 Server!2 Server!4
Average
(seconds)
66,8 73,5 67,6 90,4
Standard!Deviation 33,2 35,5 30,0 43,2
Global!Row!average 70,0 76,7
Global!Row!Standard!Deviation 34,4 37,3
Figure 10. Service Time averages and standard deviations
Efficient Taxi Pick-up Operations at Airports 8/1/2011 Page 18 of 24
Figure 11. Main conflicts degrading taxi service under observed conditions.
Figure 12. Queue Length evolution
Figure 13. In-Queue Waiting Time evolution
Efficient Taxi Pick-up Operations at Airports 8/1/2011 Page 19 of 24
Figures 12 and 13 describe the overall performance of the taxi pick-up operations observed at
Lisbon International Airport. Once the delays start during a busy period, they increase rapidly,
leading both to long queues (as many as 130 in a queue were observed) and to long wait times
(more than 20 minutes). In this connection, it should be stressed that the delays were not due to
a lack of available taxis – many were consistently available in the taxi pool about 100m away.
The delays can be attributed to the unsatisfactory design of the pick-up operations. This
phenomenon stimulated the interest in the issue.
2. Creation and calibration of simulation model
Figure 14 shows the basic configuration of the simulation model for the taxi pick-up operations for
the system as observed. Its notable feature is that it includes separate service modules for each
of the taxi positions. This contrasts with the naïve alternative of assuming that the service rate is
the same for each taxi, and that the overall system can be reasonably considered to be a single
queue. Costa implemented the model using the modular capabilities of the Simul8 package.

Figure 15 illustrates the process of calibrating the simulation model. It compares the queue
length observed with that predicted by the simulation. In this case, the simulation estimated the
maximum length of queue and the pattern of its evolution reasonably accurately, although it did
estimate that the queue would build up earlier. A more detailed analysis, beyond the scope of
this study, could investigate such phenomena. The calibration is satisfactory for the purposes of
this demonstration of the process.

3. Exploration of alternatives
This application built four alternative scenarios and tested them for performance:
• Scenario I, Increase the number of service lanes/rows and servers, consisted in enlarging
the inner-curbside area to allow for three lanes of taxis to serve simultaneously at peak
hours, instead of the current two. The modeling of this situation consisted in adding two
extra servers to the service area, with the assumption that they follow service time
distributions that are similar to those of the remaining servers.

Scenarios IIA and IIB, Different service area configuration (single lane, multiple servers),
feature an inner service lane, loading several taxis at a time, and a free lane, allowing
taxis to safely and quickly bypass front colleagues. Scenario II.A had three service
positions and Scenario II.B had four.
• Scenario III, Multiple queues and service segmentation, explored the possibility of
creating a secondary and smaller queue (and dedicated road access) with two service
positions near the exit of the inner-curbside road, on the far left end of the Terminal
facade, simultaneously keeping the current existing taxi service area functioning.
Figures 16 and 17 show the simulation results for the several tested scenarios, in terms of queue
size and queuing time indicators, respectively.

4. Choice of preferred design
Any coherent and responsible evaluation of alternatives must take into account not only the main
average indicators of service and queuing, but also extreme values as seen in Figure 18. The
combination of these two evaluation perspectives allows the analyst to weight and show the
peak-hour system behavior better.
Other interesting and important indicators can become important, especially in space or costconstrained contexts, and decisively influence the final choice of preferred design. In the case of
Lisbon, although Scenario I was consistently the best in terms of results, the airport operator
perceived it as a less safe, space demanding and expensive alternative. The decision-makers
considered Scenario IIB the preferred design, because it implied better levels of safety and
affordability at the cost of only slightly less performance.

Further Application: JetBlue Terminal at New York/Kennedy
Based on the successful application in Lisbon, we have been invited to collaborate on a similar
application to the new JetBlue Terminal at New York/Kennedy (Kamga et al, 2011). This process
appears to be working well. As of the date of submission of this paper (August 1, 2011), these
results are unavailable for release. If appropriate, we would hope to be able to include them in
the presentation version of this paper.
Conclusion
This paper proposes and demonstrates a practical solution to a long-standing issue in the design
of ground transportation services for airports. It provides airport managers with the means to
analyze taxi operations quantitatively, which has previously not been available. Based on this
approach, they should be able to investigate alternative designs for their taxi pick-up sites,
identify preferred solutions, and improve these operations that so often have been deficient.

Acknowledgements
The authors gratefully acknowledge the financial support of the MIT-Portugal Program, funded by
the Portuguese Fundação para a Ciencia e Tecnologia (FCT); and the professional assistance of
ANA (Aeroportos de Portugal, SA) and of the SIMUL8 Corporation. We also appreciate the
cooperation of colleagues, in particular Nuno Costa and Pedro Teixeira; Rosário Macário of the
Instituto Superior Técnico of Lisbon, and Regine Weston of Arup and of Weston-Wong